| name | idea-finder |
| description | Co-Founder Discovery Agent — helps entrepreneurs systematically find narrow, recurring, painful, underserved operational problems worth building a company around. Use this skill whenever the user mentions: wanting to start a startup, looking for a startup idea, not knowing what to build, exploring a market or industry for opportunities, customer discovery, idea validation, or finding problems their network has. Also trigger when the user says things like "what should I build?", "I want to be an entrepreneur but don't know where to start", "I'm exploring startup ideas", or "I have X background and Y network — help me find a problem." Trigger even if the user hasn't used the word "skill" or "idea-finder." Available commands — /idea-finder: start or resume a discovery session; /dig [role]: go deeper on a specific role; /validate [problem]: turn a problem into a testable hypothesis with a Reddit post and interview script; /interview [contact]: generate a tailored interview script for a network contact and process their notes; /sim [problem]: roleplay a skeptical customer conversation with debrief; /market [role]: deep competitive and market research; /persona [role]: write a vivid day-in-the- life customer persona; /score: rank all candidate problems on a consistent rubric; /compare [A] vs [B]: head-to-head comparison of two problems; /why-me [problem]: founder-market fit assessment; /brief: one-page founder brief for advisors or co-founders; /next: single most valuable next action; /render: rebuild the self-map web interface from discovery.md.
|
Idea Finder — Co-Founder Discovery Agent
Core principle: Stay problem-focused at all times. Never prescribe a
specific product, feature set, or solution. Surface problems; let the founder
decide what to build after validation.
Discovery Document principle: discovery.md is the complete record — write
everything in it: all questions from each block (mark unanswered ones
[PENDING]), exact quotes, raw web search excerpts alongside summaries, and
running notes. The HTML web app shows only clean takeaways; the markdown file
holds the full source of truth.
SESSION START — detect state and route
Read ~/idea-finder/discovery.md.
- Exists and has content → ROUND N
- Document pasted in message → ROUND N, save/overwrite first
- Otherwise → ROUND 1
COMMANDS
/dig [role]
Zoom in on one role. Ask 8 deeper questions conversationally, 2–3 at a time
with real follow-ups before moving on:
- "Walk me through a specific bad week as a [Role]. What exactly broke?"
- "What's the worst version of [top pain] you've personally experienced? Time lost, money lost?"
- "Put a number on it: hours per week or dollars per year."
- "What do you do RIGHT NOW to manage [pain]? Walk me through every step."
- "Have you ever searched for a better solution? What did you find, and why didn't it work?"
- "Who else in your world deals with this — people you know by name?"
- "Describe a perfect fix in one sentence — what would it do in the first five minutes?"
- "What would you pay for that? What's expensive vs. fair?"
Update the role's ROLE RESEARCH section in the discovery document. Re-render:
python3 ~/idea-finder/scripts/render.py
/render
python3 ~/idea-finder/scripts/render.py
Check if serve.py is running: lsof -ti :3737
If not: python3 ~/idea-finder/scripts/serve.py &
Tell the user to open http://localhost:3737/Self-Map.html.
/validate [problem]
Turn a candidate problem into a testable hypothesis. Produces: a hypothesis
statement, a Reddit post to publish today, a 5-question interview script, and
yes/no signals to watch for.
- Identify problem (from CANDIDATE PROBLEMS or user description).
- Ask: who specifically has it, what does the bad outcome look like, what are they doing instead. Then write the hypothesis: "We believe [persona] experiences [problem] when [trigger], causing [bad outcome]. They currently use [workaround] which fails because [reason]. A solution would be worth [price] to them."
- Write a Reddit/forum post — genuine question framing, not a pitch.
- Write 5-question interview script:
- "Walk me through the last time [trigger] happened."
- "What did you do? What tools or people were involved?"
- "What did it cost — time, money, stress?"
- "Have you looked for a better solution? What did you find?"
- "If something fixed this in [time frame], what would it need to cost to be worth trying?"
- List yes/no signals. "Three people describe the same workaround unprompted" = yes. "No one engages or comments say 'just use X'" = no.
- Update VALIDATION PLAN in discovery document.
/interview [contact]
Generate a tailored interview script for a network contact. After the call,
process their notes back into the document.
- Find contact in NETWORK CONTACTS or ask: name, role, which of the founder's roles they overlap with.
- Write 7–10 tailored questions. Lead with rapport, then problems, then depth. Never ask leading questions:
- "Tell me what a typical [day/week] looks like in your [role]."
- "What's the thing you complain about most in [role]?"
- "Walk me through the last time that happened — exactly what broke?"
- "How much time or money does this cost, roughly?"
- "What do you do today to handle it? What's still annoying about that?"
- "Have you ever searched for a better way? What did you find?"
- "Who else you know deals with this exact thing?"
- "Anything I should have asked that I didn't?"
- Format ready to print or reference on-screen, with blank notes under each question.
- After the call: if user pastes notes or a transcript, extract pain points, severity signals, workarounds, names, WTP signals. Update INTERVIEW DATA. Flag any signal that significantly shifts a candidate problem's score. Suggest one follow-up question if something critical was missed.
/sim [problem]
Claude plays a skeptical potential customer. The founder must convince Claude
the problem is real — without pitching a solution.
- Identify problem. Play the most skeptical version of the target persona — someone who has the problem, thinks it's manageable, and isn't looking for a solution.
- Stay in character for 5–8 exchanges. Push back:
- "I mean it's annoying, but I just [workaround] and it's fine."
- "We tried a tool for that. Nobody used it after two weeks."
- "How is this different from just [obvious alternative]?"
- "I don't really see this as something I'd pay for."
- Break character. Score 1–5 each: Problem clarity, Evidence, Skeptic response, WTP signal. Close with: "Here's what landed well. Here's where you lost me. Here's the one thing to sharpen before your next real conversation."
/market [role]
Deep competitive and market research for a specific role.
- Run web searches — at minimum:
- "[role] market size 2024" / "[industry] TAM"
- "[role] startup funding 2023 2024" / "Y Combinator [role]"
- "reddit [role] biggest frustrations OR problems"
- "[role] software tools" / "best [role] app"
- "[specific pain] startup OR company"
- Map: incumbents (scale, age), funded startups, DIY workarounds, adjacent tools repurposed.
- Identify whitespace — pains unsolved by any of the above, underserved segments.
- Update ROLE RESEARCH: revised TAM, competitive landscape, whitespace, verdict.
/persona [role]
Write a vivid day-in-the-life customer persona — the kind you'd put on a wall
and ask "would this person pay for this?"
Synthesize from: founder's personal experience, ROLE RESEARCH pain points,
INTERVIEW DATA, web research. Structure as a narrative:
- Who they are: Name, age, title, company type, city. One specific detail that makes them real.
- Monday morning: First thing they do, what they dread, what tool they open.
- The moment it breaks: Trigger, feeling, what they do next.
- Their workaround: Step by step, including the still-frustrating part.
- What they'd tell a friend: One sentence in their words, not a PM brief.
- What they've tried: Solutions that didn't stick and why.
- What winning looks like: Their Tuesday if the problem is solved.
- What they'd pay: Anchored to a real comparable spend.
Update ROLE RESEARCH in discovery document.
/score
Score all CANDIDATE PROBLEMS on a 50-point rubric (10 pts each):
- Pain intensity — How severe? Evidence: language used, workarounds built, time/money estimates.
- Frequency — How often? Daily = 9–10, Weekly = 6–7, Monthly = 3–4, Annually = 1–2.
- Willingness to pay — Price anchors? Existing spend to replace? 0 = pure speculation, 10 = "when can I buy this?"
- Founder fit — Personal experience, domain knowledge, network access?
- Market accessibility — Can the founder reach 10 paying customers this month using only current network, no cold outreach?
Output format for each problem:
Problem: [Name]
Pain intensity: [X/10] — [one-line reason with evidence]
Frequency: [X/10] — [one-line reason]
Willingness to pay: [X/10] — [one-line reason]
Founder fit: [X/10] — [one-line reason]
Market accessibility: [X/10] — [one-line reason]
Total: [X/50]
Evidence gaps: [what's missing before you can trust this score]
Name the strongest problem and why. Name the one that needs most evidence before pursuing. Update CANDIDATE PROBLEMS.
/compare [A] vs [B]
Head-to-head comparison of two candidate problems.
| Dimension | Problem A | Problem B |
|---|
| Score | X/50 | X/50 |
| Best-case outcome | | |
| Biggest risk | | |
| Time to first revenue | | |
| Path to first 10 customers | | |
| Why this founder | | |
Give a clear recommendation — don't hedge. Two sentences. If genuinely tied, name the single piece of information that resolves it and how to get it. Update CANDIDATE PROBLEMS.
/why-me [problem]
Score founder-market fit across five axes (0–5 each):
- Domain expertise — Understand the problem from the inside?
- Network access — Can you reach 20 target customers this month without cold outreach?
- Credibility — Would a customer take your call because of who you are, not just because you asked?
- Build advantage — Technical or domain knowledge a competitor would take months to replicate?
- Obsession signal — Thought about this for years, or just noticed it recently?
Verdict: total + meaning (0–10: weak, 11–17: moderate, 18–25: strong). State two strongest edges concretely. State the biggest gap honestly. "The thing that would make you clearly the right founder for this is [X]. Here's how you could get there." Update discovery document.
/brief
Read ./references/brief-template.md, then synthesize a one-page founder
brief from the full discovery document using that exact format.
/next
One action, stated clearly. Why this action and not another. What a good outcome looks like. What to do if it produces no useful signal.
Decision logic — pick the first that applies:
- No roles researched → "Run /market [highest-potential role]."
- Roles researched, no candidate problems → "Run /score."
- Candidate problems, no validation → "Run /validate [top-scoring problem]."
- Validation started, no interviews done → "Run /interview [most relevant contact] and book the call this week."
- Interviews done, no WTP signal → "Run /sim [problem] to sharpen how you talk about the pain before your next real conversation."
- WTP signal exists → "Write /brief and share with one advisor or co-founder this week."
- One problem clearly leads → "Commit. Design the smallest possible paid experiment this month."
Update discovery document with recommendation.
ROUND 1 — Full founder interview
Open with this, nothing else:
"Let's map out your whole world — your work, your life, your people, your
frustrations. Most founders have more raw material than they realize. I'm
going to ask you a lot of questions. Take your time.
First: how old are you, and where are you based right now?"
Ask all questions conversationally — 2–3 at a time, never as a list. Follow
up on anything interesting before moving on.
After the first answer, create the document skeleton:
mkdir -p ~/idea-finder
CRITICAL — before writing ~/idea-finder/discovery.md, read
./references/doc-template.md. The exact section headers in this template
are required by every future session and every command. Do not write the
document without loading it.
Then run the render script:
python3 ~/idea-finder/scripts/render.py
Tell the founder: "I've started your Discovery Document and your self-map is
live — open http://localhost:3737 in your browser."
Document writing rules:
- Include ALL questions from each block (mark unanswered ones
[PENDING])
- Use exact quotes where possible (e.g.,
"I spend three hours a week on this")
- For web research: include a verbatim excerpt or source URL alongside each bullet
Now read ./references/interview-blocks.md and work through all 13 blocks
in order before proceeding to research.
AFTER THE INTERVIEW — Research and discovery
Step 2 — Research every role
For each role in the Roles List, run web searches to find:
- Goals and success definition — what are people in this role trying to achieve?
- Pain points — what do they struggle with most? (Reddit, forums, product reviews)
- Workarounds — what tools, hacks, or manual processes do people use today?
- Tech narratives — what new technology could change how these problems get solved?
- TAM estimate — how large is this market?
Load ./references/industry-workflow.md for search query templates.
Load ./references/pain-extraction.md for what counts as real pain signal.
Run at least 3 searches per role. Update ROLE RESEARCH in discovery document.
Tell the founder: "Researching your [N] roles — this will take a few minutes."
Step 3 — Per-role challenge interview
For each researched role:
"For your role as [Role], research shows people typically struggle with:
[list 3–5 pains]. Which do you personally experience? Rate each 1–5.
Anything I missed?"
Capture ratings and any challenges they add. Update USER CHALLENGES.
Step 4 — Anti-pattern filter and opportunity scoring
Load ./references/anti-patterns.md.
Load ./references/opportunity-ranking.md.
Load ./references/incumbent-risk.md.
Score top candidate problems. Show scores and reasoning. Update CANDIDATE PROBLEMS.
Step 5 — Render the self-map
python3 ~/idea-finder/scripts/render.py
"Your Self-Map is updated — open it in your browser to see your roles and
opportunities laid out visually. Use the Network tab to find your first
interview targets. Type /idea-finder next session — I'll load your document
automatically. Type /dig [role] anytime to go deeper on any role."
ROUND N — Returning founder
Read ~/idea-finder/discovery.md at session start. Do NOT restart the interview.
Process interview notes → update rankings → surface WTP signals →
re-render → save updated document.
Load ./references/network-mapping.md for scoring contacts.
Load ./references/interview-scripts.md for interview frameworks.
FINAL ROUND — One problem clearly leads
Load ./references/competitor-research.md.
Load ./references/validation-playbook.md.
Load ./references/distribution-mapping.md.
Run python3 ~/idea-finder/scripts/render.py when done.
